analiza aiqbench
This research exposes a critical flaw in bias evaluation by revealing that large language models often tokenize names unevenly, granting direct single-token access to some while fragmenting others based on demographic metadata. The authors introduce NameTrace, a novel framework that measures how these lexical disparities persist as internal representations and systematically skew outcomes in high-stakes domains like hiring and lending. By proving that unequal input representation directly drives biased behavior, the study underscores the necessity of addressing vocabulary-level inequities before they manifest in model outputs.
Who Gets a Token, and What Does It Carry? Unequal Name Support and Concept Access in Large Language Models
arXiv:2609.34065v1 Announce Type: cross Abstract: Names are personal identifiers, but they also carry social meaning and are widely used to evaluate how language models treat different people. Such evaluations typically assume that matched names are comparable model inputs. We show that this assumption often fails at the lexical interface: matched names are not necessarily matched inputs. Some names receive direct single-token access, while others are assembled from multiple subwords, creating unequal name-surface support. Across nearly half a million first names and 12 LLM-associated tokenizers, direct lexical access is highly selective, model dependent, and uneven across race- and gender-associated name metadata. We introduce NameTrace, a model-native, fine-grained, pre-behavioral framework for measuring whether unequal name-surface support remains a vocabulary property or becomes visible in task-relevant internal representations. NameTrace measures concept accessibility from the model's own probabilities over task-specific adjective axes with continuous task-aligned weights. On matched atomic and short-fragmented names within the same race/ethnicity--gender-associated strata, support predicts systematic differences in concept accessibility across fellowship, hiring, clinical assessment, and lending. These differences persist across all eight matched strata, extend across model families, and transfer to unseen names. Hidden-state interventions further show that the measured task directions have downstream leverage, shifting later constrained choices. Unequal lexical support is therefore demographically structured at the input and remains visible in task-relevant model computation. NameTrace makes lexical comparability measurable, supporting a broader principle: behavioral comparability begins with lexical comparability.
Abstract:Names are personal identifiers, but they also carry social meaning and are widely used to evaluate how language models treat different people. Such evaluations typically assume that matched names are comparable model inputs. We show that this assumption often fails at the lexical interface: matched names are not necessarily matched inputs. Some names receive direct single-token access, while others are assembled from multiple subwords, creating unequal name-surface support. Across nearly half a million first names and 12 LLM-associated tokenizers, direct lexical access is highly selective, model dependent, and uneven across race- and gender-associated name metadata. We introduce NameTrace, a model-native, fine-grained, pre-behavioral framework for measuring whether unequal name-surface support remains a vocabulary property or becomes visible in task-relevant internal representations. NameTrace measures concept accessibility from the model's own probabilities over task-specific adjective axes with continuous task-aligned weights. On matched atomic and short-fragmented names within the same race/ethnicity--gender-associated strata, support predicts systematic differences in concept accessibility across fellowship, hiring, clinical assessment, and lending. These differences persist across all eight matched strata, extend across model families, and transfer to unseen names. Hidden-state interventions further show that the measured task directions have downstream leverage, shifting later constrained choices. Unequal lexical support is therefore demographically structured at the input and remains visible in task-relevant model computation. NameTrace makes lexical comparability measurable, supporting a broader principle: behavioral comparability begins with lexical comparability.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Emerging Technologies (cs.ET); Machine Learning (cs.LG)
arXiv:2609.34065 [cs.CL]
(or arXiv:2609.34065v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.34065 arXiv-issued DOI via DataCite (pending registration)
Text integral sub licența deschisă a publicației · 259 cuvinte
Subiecte conexe
The J-Space Debate, Agent Swarms, and Pacing Frontier AI - Digital Minds Newsletter #4
LessWrong · 18 sept. 2026
Multi-Vector (Late Interaction) Embedding Models with Sentence Transformers
Hugging Face · 18 aug. 2026
Image Generation Models: Choosing One, Fixing Errors
OpenRouter · 27 iul. 2026
[AINews] not much happened today
Latent Space · 3 oct. 2026